This isn't complete. It omits an important detail that has never appeared in US open publications but has appeared in some materials from the former USSR.
What this looks like is close to what Klaus Fuchs gave to the Russians when he was spying at Los Alamos. A similar rough sketch was published decades ago, but not one with dimensions.
There's not much new here. If you're familiar with CALEA, the law that hooked the Government into the phone system big-time, this is basically the same set of requirements the FBI wanted for voice calls. There was a big disagreement in the voice world over in-band signalling. The question was whether a "pen register" warrant authorized access to signalling data that goes over the voice channel, like Touch-Tone tones sent to some non-carrier device. The FBI was bitching about that for years.
The trouble with all this stuff is that Congress didn't mandate proper auditing. Every surveillance event in CALEA ought to be logged by the Judicial Branch, at the Administrative Office of the U.S. Courts. We don't have that.
So basically you want the windows registry for LINUX!?!?
The Windows registry is not a database. It's an unstructured tree. There are no indices, no tables, no record structures, and no locking. You can't look up anything, other than by brute force.
This was written by someone from the "I'm so l33t because I can do system administration" crowd. Or worse, "I'm so l33t because I can type command line commands". (Visualize a fat guy in a "Got Root" T-shirt.) Wake up, people. It's 2008, not 1978.
If a system needs much administration, it's badly designed. You should never have to tell the computer something it already knows. Everything that needs "tuning" should self-tune. That's the way of progress. TVs through the 1960s had dozens of screwdriver adjustments. Todays's TVs have none. That's progress. When was the last time you saw a vertical hold control?
One of the strengths of the original Mac was that it didn't have a command line. This forced designers to think through these issues and solve the problems, instead of shoving them back in the user's face. (The original MacOS had serious problems, but they came mostly from the fact that, down at the bottom, it was an OS at about the level of the DOS resident, with no process management. It was supposed to fit in 64K, remember, and it did fit in 128K, painfully.)
We really should be doing Linux/UNIX system administration with a real database (maybe SQLite) at the bottom, rather than text files. Many of the system administration troubles with UNIX and Linux come from trying to use text files to do a database's job. With a database, you get consistency checks, locking, security controls, standardized record structure, and indexing. Without one, you get unreliable hacks (lock files, "vipw", all-or-nothing security, and the messes inside DNS and Sendmail).
Once you've gotten away from text files, it's much easier to do automated and remote adminsitration, or to put a real GUI client on something. Too many UNIX GUI clients are kludges on top of a command line program. You see this when the GUI program doesn't really understand what's coming back from the lower level, and just blithers low-level text at the user.
A few years ago, I was visiting a small PC manufacturer. They were trying for product differentiation from Dell, HP, etc., and had a row of "concept cases" on display. There was one with Viking horns. One like a Darth Vader mask. One something like this one. One that looked like a 1940s Telefunken radio. Some of these went into production. If you really want a PC that looks like a yellow Samurai mask in plastic, they have some in stock.
I saw one of the Viking horn models in a surplus store recently.
Very nice work. These antenna designers have the enormous advantage of having antenna simulation software that runs on PCs. It's a problem where intuition isn't good enough, hand calculation involves oversimplification, and repeatable experimental work requires either an RF anechoic chamber or a big flat field in an RF-quiet area where you can transmit on the band in question. Antenna test ranges thus tend to be located in Outer Nowhere, and hams who do antenna development usually go to some isolated place to test. Otherwise, you can't tell if a small change made things better or worse.
It's not funny any more. Not since the US Army started deploying their killbots in Iraq and Afghanistan.
The moment when mobile robotics got serious can be pinpointed. It was on the first day of the 2005 DARPA Grand Challenge. Now, remember that the 2004 Grand Challenge had been a total joke. None of the vehicles really worked. Some of them crashed in the starting gate. Some of them rolled over. None made it more than a few miles along an easy course. The whole thing was just embarrassing to everyone involved.
Then, at the 2005 Grand Challenge, 43 robot cars showed up at the California Motor Speedway and they all worked. The non-starters had been eliminated in earlier rounds. Suddenly it wasn't a joke any more.
It's too late for wedgebots on the Comedy Channel. We're past that.
Nuclear reactor pressure vessels are a real problem. Most of the larger ones are in fact built up from welded sections. This isn't an easy welding job, and inspection of welds is a big headache. Several Japanese nuclear plants have had problems with cracks in pressure vessel welds, although in internal reactor components welded to the shell, not the shell itself. So making the pressure vessel and its internal support structures from one big forging makes a better product.
The environment of a reactor pressure vessel is tough. First, there's "embrittlement". Neutrons are constantly blasting apart the atoms in the pressure vessel, and over a period of years, this structural damage adds up. Then there's corrosion. There have been major corrosion problems requiring reactor shutdowns from carbon dioxide and boric acid corrosion inside the pressure vessel. Remember, this is a steam pressure vessel; at steam temperatures and pressures, minor corrosive effects at room temperature become big problems.
High quality welding of thick steel sections is a tough problem. Many approaches have been tried. The general idea is to make a V-shaped notch and fill it in during the welding process. Doing this in a way that's no weaker than the surrounding material is hard.
Electric arc welding under an inert gas is the usual approach. Electron beam welding and laser welding have been tried. Then there's the problem of approach angle - welding on a vertical surface is not easy. Quality control requires X-rays, ultrasonic tests, and regulators that aren't corrupt.
So there's much to be said for building the pressure vessel as one big forging. Of course, then there's the problem of delivering a 550-ton object to the job site. There are companies that can do that, if you can find them a clear path from a seaport.
Sword making technology is relevant to the making of big forgings. Swords are built-up forgings. This is unusual in modern metalworking; most modern forged objects, like tools, are banged out in one piece by equipment much larger than the thing being manufactured. Big pressure vessels are built-up forgings; the scale requires it. In Japan, it's considered a good doctoral thesis in metallurgy to improve on sword making technology. So smart people are still thinking about the technology of built-up forgings. Nobody else bothers much.
This has been my main criticism of "p2p" user-level networking for years. The selection of "peers" has no clue about network structure. The routing performance is just awful.
Finally, someone is doing something about it.
One problem is that, from an endpoint perspective, it's tough to extract network topology and bandwidth. Hop count is only moderately useful. But there are a few tricks one can use.
There are several basic numbers of interest - bandwidth, delay ("lag"), hops,"bottleneck points" and
commercial boundary crossings. Each of these can be measured.
Delay, or lag, is the easiest to measure. A few pings and you've got it.
With bittorrent, you're not committed to staying with a peer for an entire download. So you can observe the bandwidth of the peers you're talking to and preferentially use the higher bandwidth ones. You really have to transmit for a while to get a solid bandwidth number, especially since Comcast introduced "Boost" quality of service, which increases bandwidth allocation for a few seconds on demand, then reduces it.
If you do a traceroute, you'll usually observe that many hops show low lag (those are usually hops within a single data center) while others show higher lag. The number of high-lag hops is the number of "bottleneck points" in the path.
Commercial boundary crossings occur then packets cross from one ISP to another at a peering point. Users don't notice this much, but carriers are very interested in minimizing that traffic. Converting IP addresses to autonomous system numbers, as someone mentioned, can tell you when you're crossing a boundary.
So it's possible to collect enough data to do intelligent routing without much help from the network provider. What to do with that data is a separate question, but a solveable one.
Last year, I heard a VP from Applied Materials give a talk on their solar panel operation.
Applied Materials is a big, profitable company that makes a big fraction of the world's semiconductor and flat panel fab gear.
Key points:
From their perspective as a semiconductor wafer fab equipment builder, this looks like a nice business. Their costs are going down, and the competition (oil, gas, etc.) has costs that are going up. The market is nowhere near saturation. They see big profits in the near future.
Charts of costs per watt vs time show a steady decline, like most other things in semiconductors.
Their costs fell below other energy sources in very sunny areas around 2006-2007.
Half the installed cost of a solar system is installation. They need better technology at
that end than "a guy with a pickup truck". They're working on panels that form roof, wall, or window sections, rather than just being bolt-ons.
Applied Materials is ready to build a "gigawatt fab", one that makes a gigawatt worth of panels a year. (One such fab could build enough panels to power most of Southern California's air conditioning load in a decade.)
Their solar technology is derived from their flat-panel display technology, where they make five square meters of panel at a time.
Applied Materials has much better quality control than many solar-only companies, because their technology is derived from IC and display fab, where the allowed defect level is very low. Their whole production process is heavily automated and monitored under tight software control, using Applied Materials software and sensors from semiconductor fab control.
It takes two years worth of energy output to pay back the energy used to make an Applied Materials solar panel.
They think this can be brought down to six months worth of energy.
They bought a "roll to roll" process company because they think that approach might eventually be cheaper, but for now, the flat-panel like fab is better. They see R&D as steady process improvement, as with semiconductors. If somebody develops a breakthrough technology, they'll buy or license it and make it work in volume. If not, they'll continue to improve their processes.
Their business goal is to have 75% of the world's solar panels made by Applied Materials machinery.
This was a big-company manufacturing executive talking. He never mentioned "green" or "eco" anything; he focused on volume and profitability. That's encouraging. This is finally happening for real.
I'm still running Windows 2000 and Word 97. The later Microsoft OSs are worse. windows 2000 was the last Microsoft product where the user was in charge. After that, they're all slaves to the mothership in Redmond, with a steady stream of updates being pushed at you, some of them hostile.
Actually, this is because I'm not developing for the Windows desktop. My real work in the last few years has either been on Linux servers or real-time QNX machines. So the desktop machine just doesn't matter that much.
I have a current Firefox, OpenOffice, Python, Java, etc. I just don't see any need to buy any of Microsoft's products.
The Second Coming of Jesus Christ is clearly the most significant vapor promise that never got delivered. The marketing organization has been promoting it for almost two thousand years and they still haven't delivered.
It's taken an incredibly long time for something better to replace the M1911A1, the "45 auto". Glock finally did it. The big problem with the M1911A1 was a bad safety design; even after the magazine is removed, there can still be a round ready to fire in the gun.
Col. David Hackworth once headed a study for a M1911A1 replacement for the Army. They discovered that, over the life of the product, it had killed or injured more US Army soldiers through accidents than enemy by intent. This reflects the military use of pistols; anybody with a handgun isn't expecting trouble. If you're headed into a fight, you bring something heavier. So handguns are carried by flyers, tankers, officers, MPs, and others who don't use them much and whose job isn't focused on the gun. Thus, accidents.
I did some Google searches, hoping to find some historical info on NSA's cryogenic computing efforts, and found this, a 2005 plan out of NSA to build a 50-100GHz computer by 2010.
They want faster CPUs, not more CPUs. The commercial world isn't even
trying any more. After reading this paper, one can see why.
By throwing a few hundred million, and liquid helium, at the problem, they might get a 20x performance
gain over commercial microprocessors. The CPU has to run at 4 degrees Kelvin, liquid helium temperature. And it has to be kept at 4K while dissipating about a kilowatt.
The technology is totally nonstandard. The basic components are Rapid Single Flux Quantum devices running at 4K. The logic voltage power voltage is 3-5 mV.
Signals are around 200 microvolts. This stuff requires custom semiconductor fabs to make.
Getting data out of the low-temperature zone is a very tough problem,
and optical interconnects have to be used. The proposed memory bandwidth
is huge: "For example, a particular architecture may require half a million data
streams at 50 Gbps each between the superconducting processors
and room-temperature SRAM." Developing devices to drive the output
data links from the low temperature zone, without causing too much heating in
the cold part of the system, is a big part of the problem.
The justification for all this is in Appendix E, and sounds totally bogus.
Either there's some desperate need for this technology they don't mention, or
it's a boondoggle. There must be something important for which parallelism won't work.
It's surprising to see this from NSA, because most signal analysis and crypto problems
parallelize well.
It's a neat little museum. Everything there is familiar to people in the field, but it's nice to see the actual hardware.
I would have liked to see hardware from the NSA/IBM foray into cryogenic computing. NSA funded a long effort from 1960 or so to build a 1GHz computer, decades before anybody else. ("I want a thousand megacycle machine! I'll get you the money" - NSA director) IBM
developed components that ran in liquid nitrogen. Apparently some special purpose hardware was built using this technology, but not a full-scale computer. The components were too big (each gate required a tiny coil) and ICs won out.
SIGSALY is a reminder of just how hard it was to do anything with WWII electronics. SIGSALY is straightforward; it's a speech encoder and digitizer fed through a one-time key system. The keys were stored on phonograph records, made in pairs and shipped in advance. This was VoIP, version 0.000001. The system thing took 40 racks at each end, and a staff of fifteen at each site to keep it running. The record turntables had to be mechanically synched; there was at that time no memory device suitable for storing even a modest portion of the of key so that the thing could be synchronized electronically. There was no clock sent on the data channel; synchronization was entirely manual. Unclear why they did it that way. The display at NSA is a mockup.
Bletchley Park in the UK is also worth a visit. Go on a weekend when the volunteers show up; the weekday guides don't know much about the technology.
Right now, it looks like the site is being moved. The name "ratemycop.com" is registered with "name.com", not GoDaddy. GoDaddy was providing hosting only. So moving it to another
server is easy.
Checking with the authoritative name server for the domain (NS1.MYCPANELHOST.INFO), we get back
[205.234.222.18] as the IP address. That's actually "mycpanelhost.info", indicating this is a site using named virtual hosting (many domains on the same IP address). So addressing the site by IP address just gets you a default "Welcome to Apache" page.
The new IP address hasn't propagated through DNS yet. My local DNS is returning "Addresses: 72.167.159.53, 205.234.222.18". That 72.167.159.53 address is the old GoDaddy address. There's a 7 day TTL on the DNS entry, with 6 days 5 hours to go, so it may take a while for the DNS system to purge the GoDaddy address worldwide. Some users are seeing the new site; some are seeing the old GoDaddy page.
GoDaddy is already out of the picture and has no control over the site.
We're just waiting for DNS propagation, after which the new site should be visible everywhere.
Why don't these black boxes stream their data live to satellites during the entire trip?
There are privacy issues. The voice data logs are normally erased after a successful flight.
Many aircraft do in fact send some maintenance data back to HQ over a data link. The current system is 2400 baud, so not much data is sent. Nor is it sent continuously. ARINC charges for receiving that data through their network of ground stations, and the cost per bit for this 1980s technology is quite high.
I would be excited... if there were more details convincing me this is a 'breakthrough.' That word gets thrown around a lot these days.
If the announcement came out of some startup, it would be questionable, but it came from General Electric Research in Schenectady, NY. That's an organization over a century old, and a big chunk of the electrical industry was invented there.
If they say they have a production process for making something in quantity, they probably do.
Jimbo Wales has a problem. He's famous, but he's not rich. Wikia is something of a dud; it was supposed to be a "commercial Wikipedia" and a "human powered search engine", but in reality it's just a free hosting service for fancruft. Wikia has ads, but the user demographic lives in their parents basement. There are paid ads for "Star Wars Encyclopedia", "We Sell Yu-Gi-Oh cards", "Free Star Wars TIE crawler", and "Free Lego Star Wars Destroyer" (this last from Overstock.com; somebody must have a warehouse full of the things.) Wikia is not making big bucks that way.
So the notion of tapping into all that Wikipedia content and having 50,000 editors slaving away to make Jimbo rich looks tempting. With some real revenue, Wikipedia could afford nice fat salaries and travel allowances for its nonprofit directors. The people at the top of some of the big nonprofits make out very well.
But it won't work. Add ads, and the better editors will bail out. Wikipedia needs an army of editors to push back the incoming tide of dreck that's added every day. Without extensive, ongoing attention, or much tighter restrictions on who can edit, Wikipedia will turn to mush. Few will do thankless jobs like recent change patrol to enrich someone else.
Most of the bad ideas in multiprocessing have already been explored in supercomputers. From the nCube to the Connection Machine to the BBN Butterfly, we have a good idea of what doesn't work.
We know three things that work - clusters, symmetrical shared memory multiprocessors. and highly parallel graphics-type engines. Everything else that's been tried, from hypercubes to perfect shuffle machines, has been a dud.
Clusters make the computing world go round. All big server farms are clusters of relatively independent machines communicating over I/O channels. "Web services" are provided by clusters. So we know that works. The job of the server designer is to make clusters cheaper, smaller, and less power-hungry. There's a ready market for hardware that does that. With Google building data centers in former aluminum smelter locations just to get cheap power, there's no question that this is a very real problem.
Machines within a cluster can be symmetrical multiprocessors. That works fine.
Asymmetrical multiprocessors are usually a pain. There's a long history of that idea in large computers, and they've consistently been disappointing.
In clusters, each CPU has plenty of memory and its own private disks. Intercommunication is limited and slow, yet not usually the bottleneck. There are faster interconnection schemes, like Infiniband, but most clusters stick with some form of Ethernet.
It's worth noting that while the Cell gets attention, the XBox 360 is more successful, and it's quite conventional. It's a 3-CPU shared memory symmetrical multiprocessor (PowerPC), with a conventional GPU (nVidia) on the back end. On the Cell, brilliant people (I know some of them) struggle to cram problems into the architecture . On the XBox, game designers develop games. Weird architecture hurts your time to market.
The real choke points today are CPU-to-memory and memory-to-disk. We may see more memory move to the CPU chip, in the form of larger caches. We may see machines
with a modest number of cores and main memory on the CPU chip. This is easy to design and improves memory access times. When a gigabyte or so can be crammed onto the CPU chip, this will look like a good option for desktop machines. One big chip will be the whole computer. That's the low-end PC of the near future.
As non-volatile memory, ("flash" and its friends) becomes cheaper, we'll face a new architectural challenge. To date, such memory has usually been treated as a fast disk. But this is suboptimal. Flash is near random access, but we're not using it that way. We need a new way to talk to flash memory, something that has file system like protection, doesn't require OS intervention, and has finer granularity than disk blocks. An interesting concept would be a flash memory/CPU combo optimized for running SQL-type databases.
Slashdot Mirror
This isn't complete. It omits an important detail that has never appeared in US open publications but has appeared in some materials from the former USSR.
What this looks like is close to what Klaus Fuchs gave to the Russians when he was spying at Los Alamos. A similar rough sketch was published decades ago, but not one with dimensions.
Looks like the open source world's answer to Microsoft Barney for Windows.
We need to do better than this.
There's not much new here. If you're familiar with CALEA, the law that hooked the Government into the phone system big-time, this is basically the same set of requirements the FBI wanted for voice calls. There was a big disagreement in the voice world over in-band signalling. The question was whether a "pen register" warrant authorized access to signalling data that goes over the voice channel, like Touch-Tone tones sent to some non-carrier device. The FBI was bitching about that for years.
The trouble with all this stuff is that Congress didn't mandate proper auditing. Every surveillance event in CALEA ought to be logged by the Judicial Branch, at the Administrative Office of the U.S. Courts. We don't have that.
So basically you want the windows registry for LINUX!?!?
The Windows registry is not a database. It's an unstructured tree. There are no indices, no tables, no record structures, and no locking. You can't look up anything, other than by brute force.
This was written by someone from the "I'm so l33t because I can do system administration" crowd. Or worse, "I'm so l33t because I can type command line commands". (Visualize a fat guy in a "Got Root" T-shirt.) Wake up, people. It's 2008, not 1978.
If a system needs much administration, it's badly designed. You should never have to tell the computer something it already knows. Everything that needs "tuning" should self-tune. That's the way of progress. TVs through the 1960s had dozens of screwdriver adjustments. Todays's TVs have none. That's progress. When was the last time you saw a vertical hold control?
One of the strengths of the original Mac was that it didn't have a command line. This forced designers to think through these issues and solve the problems, instead of shoving them back in the user's face. (The original MacOS had serious problems, but they came mostly from the fact that, down at the bottom, it was an OS at about the level of the DOS resident, with no process management. It was supposed to fit in 64K, remember, and it did fit in 128K, painfully.)
We really should be doing Linux/UNIX system administration with a real database (maybe SQLite) at the bottom, rather than text files. Many of the system administration troubles with UNIX and Linux come from trying to use text files to do a database's job. With a database, you get consistency checks, locking, security controls, standardized record structure, and indexing. Without one, you get unreliable hacks (lock files, "vipw", all-or-nothing security, and the messes inside DNS and Sendmail).
Once you've gotten away from text files, it's much easier to do automated and remote adminsitration, or to put a real GUI client on something. Too many UNIX GUI clients are kludges on top of a command line program. You see this when the GUI program doesn't really understand what's coming back from the lower level, and just blithers low-level text at the user.
This is a supercomputer?
A few years ago, I was visiting a small PC manufacturer. They were trying for product differentiation from Dell, HP, etc., and had a row of "concept cases" on display. There was one with Viking horns. One like a Darth Vader mask. One something like this one. One that looked like a 1940s Telefunken radio. Some of these went into production. If you really want a PC that looks like a yellow Samurai mask in plastic, they have some in stock.
I saw one of the Viking horn models in a surplus store recently.
Very nice work. These antenna designers have the enormous advantage of having antenna simulation software that runs on PCs. It's a problem where intuition isn't good enough, hand calculation involves oversimplification, and repeatable experimental work requires either an RF anechoic chamber or a big flat field in an RF-quiet area where you can transmit on the band in question. Antenna test ranges thus tend to be located in Outer Nowhere, and hams who do antenna development usually go to some isolated place to test. Otherwise, you can't tell if a small change made things better or worse.
It's not funny any more. Not since the US Army started deploying their killbots in Iraq and Afghanistan.
The moment when mobile robotics got serious can be pinpointed. It was on the first day of the 2005 DARPA Grand Challenge. Now, remember that the 2004 Grand Challenge had been a total joke. None of the vehicles really worked. Some of them crashed in the starting gate. Some of them rolled over. None made it more than a few miles along an easy course. The whole thing was just embarrassing to everyone involved.
Then, at the 2005 Grand Challenge, 43 robot cars showed up at the California Motor Speedway and they all worked. The non-starters had been eliminated in earlier rounds. Suddenly it wasn't a joke any more.
It's too late for wedgebots on the Comedy Channel. We're past that.
It doesn't matter if Bush vetoes it. Under current law, there is no telecom immunity. EFF vs. AT&T goes forward.
The internals of a reactor are not welded to the "pot" itself.
Unfortunately, not always. See this drawing, especially the inset detail in "Structure of Shroud Support".
Nuclear reactor pressure vessels are a real problem. Most of the larger ones are in fact built up from welded sections. This isn't an easy welding job, and inspection of welds is a big headache. Several Japanese nuclear plants have had problems with cracks in pressure vessel welds, although in internal reactor components welded to the shell, not the shell itself. So making the pressure vessel and its internal support structures from one big forging makes a better product.
The environment of a reactor pressure vessel is tough. First, there's "embrittlement". Neutrons are constantly blasting apart the atoms in the pressure vessel, and over a period of years, this structural damage adds up. Then there's corrosion. There have been major corrosion problems requiring reactor shutdowns from carbon dioxide and boric acid corrosion inside the pressure vessel. Remember, this is a steam pressure vessel; at steam temperatures and pressures, minor corrosive effects at room temperature become big problems.
High quality welding of thick steel sections is a tough problem. Many approaches have been tried. The general idea is to make a V-shaped notch and fill it in during the welding process. Doing this in a way that's no weaker than the surrounding material is hard. Electric arc welding under an inert gas is the usual approach. Electron beam welding and laser welding have been tried. Then there's the problem of approach angle - welding on a vertical surface is not easy. Quality control requires X-rays, ultrasonic tests, and regulators that aren't corrupt.
So there's much to be said for building the pressure vessel as one big forging. Of course, then there's the problem of delivering a 550-ton object to the job site. There are companies that can do that, if you can find them a clear path from a seaport.
Sword making technology is relevant to the making of big forgings. Swords are built-up forgings. This is unusual in modern metalworking; most modern forged objects, like tools, are banged out in one piece by equipment much larger than the thing being manufactured. Big pressure vessels are built-up forgings; the scale requires it. In Japan, it's considered a good doctoral thesis in metallurgy to improve on sword making technology. So smart people are still thinking about the technology of built-up forgings. Nobody else bothers much.
Here's a US NRC fact sheet. on pressure vessels, and a similar European document.
This has been my main criticism of "p2p" user-level networking for years. The selection of "peers" has no clue about network structure. The routing performance is just awful. Finally, someone is doing something about it.
One problem is that, from an endpoint perspective, it's tough to extract network topology and bandwidth. Hop count is only moderately useful. But there are a few tricks one can use.
There are several basic numbers of interest - bandwidth, delay ("lag"), hops,"bottleneck points" and commercial boundary crossings. Each of these can be measured.
Delay, or lag, is the easiest to measure. A few pings and you've got it.
With bittorrent, you're not committed to staying with a peer for an entire download. So you can observe the bandwidth of the peers you're talking to and preferentially use the higher bandwidth ones. You really have to transmit for a while to get a solid bandwidth number, especially since Comcast introduced "Boost" quality of service, which increases bandwidth allocation for a few seconds on demand, then reduces it.
If you do a traceroute, you'll usually observe that many hops show low lag (those are usually hops within a single data center) while others show higher lag. The number of high-lag hops is the number of "bottleneck points" in the path.
Commercial boundary crossings occur then packets cross from one ISP to another at a peering point. Users don't notice this much, but carriers are very interested in minimizing that traffic. Converting IP addresses to autonomous system numbers, as someone mentioned, can tell you when you're crossing a boundary.
So it's possible to collect enough data to do intelligent routing without much help from the network provider. What to do with that data is a separate question, but a solveable one.
Like hell. Who do they think they are to collect taxes? A government? It's just the music industry. It's not like it's anything important.
By the way, is there still a "DAT tax"? Probably time to push for repeal on that. Almost all DAT tapes hold backups of business machines.
100Ghz gives you only millimeters of traversal time.
That's right. The proposed CPUs are 2mm across.
Last year, I heard a VP from Applied Materials give a talk on their solar panel operation. Applied Materials is a big, profitable company that makes a big fraction of the world's semiconductor and flat panel fab gear. Key points:
This was a big-company manufacturing executive talking. He never mentioned "green" or "eco" anything; he focused on volume and profitability. That's encouraging. This is finally happening for real.
I'm still running Windows 2000 and Word 97. The later Microsoft OSs are worse. windows 2000 was the last Microsoft product where the user was in charge. After that, they're all slaves to the mothership in Redmond, with a steady stream of updates being pushed at you, some of them hostile.
Actually, this is because I'm not developing for the Windows desktop. My real work in the last few years has either been on Linux servers or real-time QNX machines. So the desktop machine just doesn't matter that much. I have a current Firefox, OpenOffice, Python, Java, etc. I just don't see any need to buy any of Microsoft's products.
The Second Coming of Jesus Christ is clearly the most significant vapor promise that never got delivered. The marketing organization has been promoting it for almost two thousand years and they still haven't delivered.
It's taken an incredibly long time for something better to replace the M1911A1, the "45 auto". Glock finally did it. The big problem with the M1911A1 was a bad safety design; even after the magazine is removed, there can still be a round ready to fire in the gun.
Col. David Hackworth once headed a study for a M1911A1 replacement for the Army. They discovered that, over the life of the product, it had killed or injured more US Army soldiers through accidents than enemy by intent. This reflects the military use of pistols; anybody with a handgun isn't expecting trouble. If you're headed into a fight, you bring something heavier. So handguns are carried by flyers, tankers, officers, MPs, and others who don't use them much and whose job isn't focused on the gun. Thus, accidents.
I did some Google searches, hoping to find some historical info on NSA's cryogenic computing efforts, and found this, a 2005 plan out of NSA to build a 50-100GHz computer by 2010.
They want faster CPUs, not more CPUs. The commercial world isn't even trying any more. After reading this paper, one can see why. By throwing a few hundred million, and liquid helium, at the problem, they might get a 20x performance gain over commercial microprocessors. The CPU has to run at 4 degrees Kelvin, liquid helium temperature. And it has to be kept at 4K while dissipating about a kilowatt.
The technology is totally nonstandard. The basic components are Rapid Single Flux Quantum devices running at 4K. The logic voltage power voltage is 3-5 mV. Signals are around 200 microvolts. This stuff requires custom semiconductor fabs to make.
Getting data out of the low-temperature zone is a very tough problem, and optical interconnects have to be used. The proposed memory bandwidth is huge: "For example, a particular architecture may require half a million data streams at 50 Gbps each between the superconducting processors and room-temperature SRAM." Developing devices to drive the output data links from the low temperature zone, without causing too much heating in the cold part of the system, is a big part of the problem.
The justification for all this is in Appendix E, and sounds totally bogus. Either there's some desperate need for this technology they don't mention, or it's a boondoggle. There must be something important for which parallelism won't work. It's surprising to see this from NSA, because most signal analysis and crypto problems parallelize well.
It's a neat little museum. Everything there is familiar to people in the field, but it's nice to see the actual hardware.
I would have liked to see hardware from the NSA/IBM foray into cryogenic computing. NSA funded a long effort from 1960 or so to build a 1GHz computer, decades before anybody else. ("I want a thousand megacycle machine! I'll get you the money" - NSA director) IBM developed components that ran in liquid nitrogen. Apparently some special purpose hardware was built using this technology, but not a full-scale computer. The components were too big (each gate required a tiny coil) and ICs won out.
SIGSALY is a reminder of just how hard it was to do anything with WWII electronics. SIGSALY is straightforward; it's a speech encoder and digitizer fed through a one-time key system. The keys were stored on phonograph records, made in pairs and shipped in advance. This was VoIP, version 0.000001. The system thing took 40 racks at each end, and a staff of fifteen at each site to keep it running. The record turntables had to be mechanically synched; there was at that time no memory device suitable for storing even a modest portion of the of key so that the thing could be synchronized electronically. There was no clock sent on the data channel; synchronization was entirely manual. Unclear why they did it that way. The display at NSA is a mockup.
Bletchley Park in the UK is also worth a visit. Go on a weekend when the volunteers show up; the weekday guides don't know much about the technology.
Right now, it looks like the site is being moved. The name "ratemycop.com" is registered with "name.com", not GoDaddy. GoDaddy was providing hosting only. So moving it to another server is easy.
Checking with the authoritative name server for the domain (NS1.MYCPANELHOST.INFO), we get back [205.234.222.18] as the IP address. That's actually "mycpanelhost.info", indicating this is a site using named virtual hosting (many domains on the same IP address). So addressing the site by IP address just gets you a default "Welcome to Apache" page.
The new IP address hasn't propagated through DNS yet. My local DNS is returning "Addresses: 72.167.159.53, 205.234.222.18". That 72.167.159.53 address is the old GoDaddy address. There's a 7 day TTL on the DNS entry, with 6 days 5 hours to go, so it may take a while for the DNS system to purge the GoDaddy address worldwide. Some users are seeing the new site; some are seeing the old GoDaddy page.
GoDaddy is already out of the picture and has no control over the site. We're just waiting for DNS propagation, after which the new site should be visible everywhere.
Why don't these black boxes stream their data live to satellites during the entire trip?
There are privacy issues. The voice data logs are normally erased after a successful flight.
Many aircraft do in fact send some maintenance data back to HQ over a data link. The current system is 2400 baud, so not much data is sent. Nor is it sent continuously. ARINC charges for receiving that data through their network of ground stations, and the cost per bit for this 1980s technology is quite high.
I would be excited ... if there were more details convincing me this is a 'breakthrough.' That word gets thrown around a lot these days.
If the announcement came out of some startup, it would be questionable, but it came from General Electric Research in Schenectady, NY. That's an organization over a century old, and a big chunk of the electrical industry was invented there. If they say they have a production process for making something in quantity, they probably do.
Jimbo Wales has a problem. He's famous, but he's not rich. Wikia is something of a dud; it was supposed to be a "commercial Wikipedia" and a "human powered search engine", but in reality it's just a free hosting service for fancruft. Wikia has ads, but the user demographic lives in their parents basement. There are paid ads for "Star Wars Encyclopedia", "We Sell Yu-Gi-Oh cards", "Free Star Wars TIE crawler", and "Free Lego Star Wars Destroyer" (this last from Overstock.com; somebody must have a warehouse full of the things.) Wikia is not making big bucks that way.
So the notion of tapping into all that Wikipedia content and having 50,000 editors slaving away to make Jimbo rich looks tempting. With some real revenue, Wikipedia could afford nice fat salaries and travel allowances for its nonprofit directors. The people at the top of some of the big nonprofits make out very well.
But it won't work. Add ads, and the better editors will bail out. Wikipedia needs an army of editors to push back the incoming tide of dreck that's added every day. Without extensive, ongoing attention, or much tighter restrictions on who can edit, Wikipedia will turn to mush. Few will do thankless jobs like recent change patrol to enrich someone else.
Most of the bad ideas in multiprocessing have already been explored in supercomputers. From the nCube to the Connection Machine to the BBN Butterfly, we have a good idea of what doesn't work.
We know three things that work - clusters, symmetrical shared memory multiprocessors. and highly parallel graphics-type engines. Everything else that's been tried, from hypercubes to perfect shuffle machines, has been a dud.
Clusters make the computing world go round. All big server farms are clusters of relatively independent machines communicating over I/O channels. "Web services" are provided by clusters. So we know that works. The job of the server designer is to make clusters cheaper, smaller, and less power-hungry. There's a ready market for hardware that does that. With Google building data centers in former aluminum smelter locations just to get cheap power, there's no question that this is a very real problem.
Machines within a cluster can be symmetrical multiprocessors. That works fine. Asymmetrical multiprocessors are usually a pain. There's a long history of that idea in large computers, and they've consistently been disappointing. In clusters, each CPU has plenty of memory and its own private disks. Intercommunication is limited and slow, yet not usually the bottleneck. There are faster interconnection schemes, like Infiniband, but most clusters stick with some form of Ethernet.
It's worth noting that while the Cell gets attention, the XBox 360 is more successful, and it's quite conventional. It's a 3-CPU shared memory symmetrical multiprocessor (PowerPC), with a conventional GPU (nVidia) on the back end. On the Cell, brilliant people (I know some of them) struggle to cram problems into the architecture . On the XBox, game designers develop games. Weird architecture hurts your time to market.
The real choke points today are CPU-to-memory and memory-to-disk. We may see more memory move to the CPU chip, in the form of larger caches. We may see machines with a modest number of cores and main memory on the CPU chip. This is easy to design and improves memory access times. When a gigabyte or so can be crammed onto the CPU chip, this will look like a good option for desktop machines. One big chip will be the whole computer. That's the low-end PC of the near future.
As non-volatile memory, ("flash" and its friends) becomes cheaper, we'll face a new architectural challenge. To date, such memory has usually been treated as a fast disk. But this is suboptimal. Flash is near random access, but we're not using it that way. We need a new way to talk to flash memory, something that has file system like protection, doesn't require OS intervention, and has finer granularity than disk blocks. An interesting concept would be a flash memory/CPU combo optimized for running SQL-type databases.